U.S. patent number 5,740,146 [Application Number 08/735,332] was granted by the patent office on 1998-04-14 for method and apparatus for reducing noise using a plurality of recording copies.
This patent grant is currently assigned to Disney Enterprises, Inc.. Invention is credited to Ronald I. Webster.
United States Patent |
5,740,146 |
Webster |
April 14, 1998 |
Method and apparatus for reducing noise using a plurality of
recording copies
Abstract
The present invention provides a method for reducing noise using
a plurality of recording copies. The present invention produces a
master file with lower noise than the available recording copies,
and avoids the problems of losing musical content caused by prior
art pop and click removers. The system comprises a recording
playback unit, a computer system with a sound input capability, and
a high capacity storage system such as a CD recorder. In operation,
a plurality of recording copies of a single recording are played on
the playback unit. These recordings are digitized by the computer
and a separate recording file is formed for each copy of the
recording. The recording files are then synchronized. The samples
from each of the recording files are then averaged to reduce the
noise components. A variety of threshold comparison techniques can
be employed to eliminate samples and/or recording files that are
outside of a computed range for that sample based on the values of
the master, the other slave files or a combination thereof.
Inventors: |
Webster; Ronald I. (Sagaponack,
NY) |
Assignee: |
Disney Enterprises, Inc.
(Burbank, CA)
|
Family
ID: |
24955330 |
Appl.
No.: |
08/735,332 |
Filed: |
October 22, 1996 |
Current U.S.
Class: |
369/107; 360/15;
369/84; 369/85; G9B/20.004; G9B/20.014; G9B/20.063; G9B/27.012;
G9B/27.017; G9B/3.097 |
Current CPC
Class: |
G11B
3/64 (20130101); G11B 20/02 (20130101); G11B
20/10527 (20130101); G11B 20/24 (20130101); G11B
27/034 (20130101); G11B 27/10 (20130101); G11B
2220/20 (20130101); G11B 2220/218 (20130101); G11B
2220/2545 (20130101) |
Current International
Class: |
G11B
27/10 (20060101); G11B 3/64 (20060101); G11B
27/034 (20060101); G11B 27/031 (20060101); G11B
20/02 (20060101); G11B 3/00 (20060101); G11B
20/10 (20060101); G11B 20/24 (20060101); G11B
003/64 (); G11B 005/86 () |
Field of
Search: |
;369/84,85,107
;360/15 |
References Cited
[Referenced By]
U.S. Patent Documents
Primary Examiner: Nelms; David C.
Assistant Examiner: Chu; Kim-Kwok
Attorney, Agent or Firm: Hecker & Harriman
Claims
I claim:
1. An apparatus for reducing noise in a recording comprising:
a means for digitizing a plurality of copies of said recording to
form digital recording files, wherein said recording files comprise
a plurality of samples;
a synchronizing means for synchronizing said recording files;
a computing means for computing a composite value for corresponding
samples in said recording files;
a compiling means for compiling an output recording file comprised
of said composite values.
2. A method for reducing noise in a sound recording comprising the
steps of:
digitizing a plurality of copies of the sound recording to form
digital recording files, wherein the recording files comprise a
plurality of samples;
synchronizing the recording files;
computing a composite value for corresponding samples in the
recording files, and compiling an output recording file comprised
of said composite values.
3. The method of claim 2 wherein said step of computing a composite
value further comprises computing an average value for
corresponding samples in the recording files, and compiling an
output recording file comprised of said average values.
4. The method of claim 3 wherein said step of computing an average
value for corresponding samples in the recording files, and
compiling an output recording file comprised of said average values
further comprises:
computing a preliminary average value of said corresponding
samples;
computing the standard deviation of the difference between each of
said corresponding samples and the corresponding preliminary
average value;
eliminating those samples that deviate from said average value by
greater than a predetermined multiple of standard deviations;
computing said average value for the corresponding samples
excluding the eliminated samples; and
storing said average value in said output recording file.
5. The method of claim 4 wherein said step of eliminating those
samples that deviate from said average value by greater than a
predetermined multiple of standard deviations further comprises
eliminating those recording files from which a predetermined number
of samples have been eliminated.
6. The method of claim 3 wherein said step of computing an average
value for corresponding samples in the recording files, and
compiling an output recording file comprised of said average values
further comprises:
selecting one of the recording files to be a master file;
computing the standard deviation of the difference between each of
said samples and the corresponding master sample value;
eliminating those samples that deviate from said master sample
value by greater than a predetermined multiple of standard
deviations;
computing said average value for the corresponding samples
excluding the eliminated samples; and
storing said average value in said output recording file.
7. The method of claim 3 wherein said step of computing an average
value for corresponding samples in the recording files, and
compiling an output recording file comprised of said average values
further comprises:
computing a preliminary average value of said corresponding
samples;
eliminating samples that differ by greater than a predetermined
percentage from said preliminary average value;
computing said average value for the corresponding samples
excluding the eliminated samples; and
storing said average value in said output recording file.
8. The method of claim 3 wherein said step of computing an average
value for corresponding samples in the recording files, and
compiling an output recording file comprised of said average values
further comprises:
computing a preliminary average value of said corresponding
samples;
eliminating a predetermined percentage of the samples which differ
the most from said preliminary average value;
computing said average value for the corresponding samples
excluding the eliminated samples; and
storing said average value in said output recording file.
9. The method of claim 3 wherein said step of computing an average
value for corresponding samples in the recording files, and
compiling an output recording file comprised of said average values
further comprises:
computing a signal-to-noise ratio for each of the recording files
based upon predetermined signal and noise characteristics;
eliminating those files that have a signal-to-noise ratio below a
predetermined threshold level;
computing said average value for the corresponding samples
excluding the eliminated files; and
storing said average value in said output recording file.
10. The method of claim 3 wherein said step of computing an average
value for corresponding samples in the recording files, and
compiling an output recording file comprised of said average values
further comprises:
computing a weighted average for the corresponding samples in the
recording files.
11. The method of claim 10 wherein said step of computing a
weighted average further comprises varying the weights in an
inverse relationship to the number of standard deviations that each
of the corresponding samples differs from the mean.
12. The method of claim 2 wherein said step of digitizing a
plurality of copies of the sound recording to form digital
recording files further comprises:
digitizing a phonograph-type record having a left contour and a
right contour in the recording groove, wherein said left contour
track is digitized to form a left contour recording file, and said
right contour track is digitized to form a right contour recording
file;
identifying samples in said left contour and said right contour
that exceed a predetermined threshold noise level;
replacing said samples that exceed a predetermined threshold noise
level from said left contour recording file with corresponding
samples from said right contour recording file;
replacing said samples that exceed a predetermined threshold noise
level from said right contour recording file with corresponding
samples from said left contour recording file.
13. The method of claim 2 wherein said step of digitizing a
plurality of copies of the sound recording to form digital
recording files further comprises digitizing the sound recording
copies at a playback speed slower than the speed at which they are
played back for listening purposes.
14. The method of claim 2 wherein said step of synchronizing the
recording files further comprises:
selecting one of the recording files as a master file;
selecting one of the remaining recording files as a slave file;
computing the standard deviation of the differences between the
values of a group of master file samples and the values of the
corresponding samples in the slave file;
shifting the slave file in time relative to the master file;
computing the standard deviation of the difference between the
values of a group of master file samples and the values of the
corresponding samples in the slave file;
repeating the above computing a standard deviation and shifting
steps a predetermined number of times;
selecting the shift value which minimizes the standard deviation of
the difference between the values of a group of master file samples
and the values of the corresponding samples in the slave file.
15. The method of claim 2 wherein said step of digitizing a
plurality of copies of the sound recording to form digital
recording files further comprises digitizing phonograph-type
records.
16. The method of claim 15 wherein said step of digitizing a
plurality of copies of the sound recording to form recording files,
further comprises playing the sound recording in a direction other
than the direction it is played in for listening purposes.
17. The method of claim 2 wherein said step of digitizing a
plurality of copies of the sound recording to form digital
recording files further comprises digitizing recordings on
tape.
18. The method of claim 17 wherein said step of digitizing a
plurality of copies of the sound recording to form recording flies,
further comprises playing the sound recording in a direction other
than the direction it is played in for listening purposes.
19. A method for reducing noise in a sound recording comprising the
steps of:
digitizing a plurality of copies of the sound recording to form
digital recording files, wherein the recording files comprise a
plurality of samples;
synchronizing the recording files;
determining whether a predetermined percentage of a group of
corresponding samples are within a predetermined range;
calculating a mode value of said corresponding samples, and storing
said mode value in an output file if a predetermined percentage of
said corresponding samples are not within a predetermined range;
and
calculating an average value of said corresponding samples, and
storing said average value in an output file if a predetermined
percentage of said corresponding samples are within a predetermined
range.
20. The method of claim 19 wherein said steps of:
determining whether a predetermined percentage of a group of
corresponding samples are within a predetermined range;
calculating a mode value of said corresponding samples, and storing
said mode value in an output file if a predetermined percentage of
said corresponding samples are not within a predetermined range;
and
calculating an average value of said corresponding samples, and
storing said average value in an output file if a predetermined
percentage of said corresponding samples are within a predetermined
range;
are performed at predetermined intervals throughout the noise
reducing process.
21. The method of claim 19 wherein said step of digitizing a
plurality of copies of the sound recording to form digital
recording files further comprises:
digitizing a phonograph-type record having a left contour and a
right contour in the recording groove, wherein said left contour
track is digitized to form a left contour recording file, and said
right contour track is digitized to form a right contour recording
file;
identifying samples in said left contour and said right contour
that exceed a predetermined threshold noise level;
replacing said samples that exceed a predetermined threshold noise
level from said left contour recording file with corresponding
samples from said right contour recording file;
replacing said samples that exceed a predetermined threshold noise
level from said right contour recording file with corresponding
samples from said left contour recording file.
22. The method of claim 19 wherein said step of digitizing a
plurality of copies of the sound recording to form digital
recording files further comprises digitizing the sound recording
copies at a playback speed slower than the speed at which they are
played back for listening purposes.
23. The method of claim 19 wherein said step of synchronizing the
recording files further comprises:
selecting one of the recording files as a master file;
selecting one of the remaining recording files as a slave file;
computing the standard deviation of the differences between the
values of a group of master file samples and the values of the
corresponding samples in the slave file;
shifting the slave file in time relative to the master file;
computing the standard deviation of the difference between the
values of a group of master file samples and the values of the
corresponding samples in the slave file;
repeating the above computing a standard deviation and shifting
steps a predetermined number of times;
selecting the shift value which minimizes the standard deviation of
the difference between the values of a group of master file samples
and the values of the corresponding samples in the slave file.
24. The method of claim 19 wherein said step of digitizing a
plurality of copies of the sound recording to form digital
recording files further comprises digitizing phonograph-type
records.
25. The method of claim 24 wherein said step of digitizing a
plurality of copies of the sound recording to form recording files,
further comprises playing the sound recording in a direction other
than the direction it is played in for listening purposes.
26. The method of claim 19 wherein said step of digitizing a
plurality of copies of the sound recording to form digital
recording files further comprises digitizing recordings on
tape.
27. The method of claim 26 wherein said step of digitizing a
plurality of copies of the sound recording to form recording files,
further comprises playing the sound recording in a direction other
than the direction it is played in for listening purposes.
Description
BACKGROUND OF THE INVENTION
1. Field of the Invention
This invention relates to the field of reducing noise in
recordings, and more specifically to producing an improved master
recording given a plurality of recording copies.
2. Background Art
Many phonograph records and other recordings that were produced
before the advent of digital recording are now being converted to
digital recordings. This process is straightforward but requires a
"clean" copy of the original recording to be digitized. A clean
copy is a copy that does not have scratches, pops, skips, hiss and
other flaws that the listener does not want to hear. Currently
there are no good solutions to digitizing a recording when there is
no clean copy. This is most often the case with old records where
the original recordings are lost or damaged. These records are said
to contain unwanted audio noise.
To understand the sources of noise in old phonograph-type records,
it is important to know something about how they are made. In
producing phonograph records, electrical signals representative of
sounds are used to vibrate a cutting tool which cuts a groove into
a base material or so-called "lacquer" used to ultimately form a
master record disc. The so-called "lacquer" is then nickel plated.
Next, the lacquer is peeled off and the nickel plating becomes the
master. Phonograph records, typically made of polyvinyl chloride,
are then produced by using the master to imprint the grooves onto a
blank record. To play back a phonograph record a needle, called a
stylus, is placed in a groove. The rotation of the record causes
the stylus to contact contours in the grooves and vibrate
accordingly. These vibrations are amplified and thereby reproduce
the recorded sound. The friction between the stylus and the groove
contours gradually distorts the shapes of the contours and thus
degrades the sound quality of the recording. In addition to this
inherent source of noise, dust and other contaminants may become
embedded in records. This dust then changes the shape of the groove
contours in the record and produces noise. Other noise sources
include scratches in the recording medium and warping of the
recording medium caused, for example, by exposure to a heat source.
The noise may sound like pops and clicks in the recording, white
noise, or a general reduction in the clarity of the sound. A "pop"
is a high level noise across a wide range of frequencies, whereas a
"click" is a high frequency, high level noise. White noise is
random noise that has a constant energy per unit bandwidth at every
frequency in the range of interest. White noise may sound like a
background fuzzy hiss on a recording.
A method and apparatus is needed to reduce the audible noise
content of analog sound recordings, particularly phonograph type
records. Digital recording technology provides high quality low
noise recordings. One widely used medium for digital recording is
the compact disc. Compact discs store music in the form of a series
of bits. Reflected light from a laser beam is used to read the
bits, which are then converted back into analog signals to play
back the recording. Because a laser is used to read the information
from the compact disc, there is no degradation of the recording
medium caused by physical contact with the recording medium as is
the case with tape players and turntables. The highly accurate
digitization process used in recording compact discs combined with
the absence of any mechanical degradation of the recording medium
during playback are important features that have led to the
popularity of the compact disc. The high fidelity audio
reproduction afforded by compact discs has also caused widespread
dissatisfaction with the noise levels of older analog
recordings.
For many old recordings, a master recording is no longer available.
All that remains may be several copies of the recording that have
been heavily used for many years. For example, often radio station
announcers apologize for playing a low quality early jazz
recording, but explain their selection on the grounds that no
re-release exists. These old recordings are likely to have
substantial noise components caused by the accumulation of
contaminants, scratches and groove contour degradation from the
playback process. Traditional techniques to remove the pops and
clicks from records detrimentally affect the sound quality of the
recording. For example, generally click and pop removers are
designed to operate using only one copy of a recording; therefore,
they cannot restore recording content that is no longer there.
Consequently, these systems typically leave empty gaps in the
recording where the pops or clicks were. The limitations of prior
art noise reducing systems discourage record companies from
re-releasing old recordings where there is no master recording.
Thus, for old recordings where a master is not available, a system
is needed to reduce the noise in the recordings, that does not
create gaps in the recordings, and avoids the other problems of the
prior art.
References related to sound recording and recording quality
enhancement include the following:
In U.S. Pat. No. 4,410,970 Law discloses a method and apparatus for
identifying the frequency bands and the time of occurrence of
"ticks and pops" on a record. Law isolates the times and
frequencies at which ticks and pops occur on a number of records.
Law focuses on obtaining a profile of noise across several records.
In other words, Law focuses on determining which frequency bands
have the highest number ticks and pops, and the times at which the
largest number of ticks and pops occur. If the ticks and pops are
limited to a certain frequency band or a certain time interval, the
ticks and pops are results of defects in the master recording. On
the other hand, if the ticks and pops are randomly distributed,
they are due to reasons other than a defective master. For example,
they can be a result of faults in the recording material.
In U.S. Pat. No. 4,186,280 Geiseler discloses restoration of old
sound recordings. Old recordings having a musical part, such as a
singing voice, a solo instrument and a formant, are restored. The
old recording is played back with the production of an electrical
acoustic signal. The amplitude of the acoustic signal is enhanced
in frequency ranges which correspond to the formants of the musical
part.
In U.S. Pat. No. 5,132,955 Hanson discloses a method and apparatus
for synchronizing multiple CD players. Sampling frequencies are
inputted to the word clock inputs of the CD players. Each of the
disks has a header disposed at the beginning of each program track.
This header has a synchronization signal .that indicates the
position of the output digital data. The synchronization signal
thus indicates the difference between positions of the different CD
players. A controller then uses this information to synchronize the
CD players.
In U.S. Pat. No. 4,851,931 Parker discloses a method and apparatus
for converting an analog audio signal into a digital audio signal
and thereafter converting the digital audio signal into a video
signal, thereby significantly compressing the original analog audio
signal. The video signal can be written onto a video disk or a
video tape.
In U.S. Pat. No. 5,021,893 Scheffler discloses a recording
duplicating system. The analog audio signals from the master
recording are converted to digital signals and stored in a buffer.
The "duplication ratio" can be any value permitted by the recording
and reproducing machines. Set up procedures to switch from one to
another duplicating ratio are thus eliminated.
In U.S. Pat. Nos. 5,041,921; 5,365,381; and 5,418,654 Scheffler
discloses a custom album recording system. A master library has a
number of recorded items from sources such as phonograph records,
tapes, and compact discs. Each recorded item has its own address.
An operator selects certain recorded items by selecting their
addresses. The selected items are written onto a large capacity
memory and from the large capacity memory are written at a high
speed onto an album size medium, such as a cassette tape. A
computer provides both analog-to-digital and digital-to-analog
conversions as needed.
In U.S. Pat. No. 4,410,917 Newdoll discloses a method and apparatus
for recording information from a master medium onto a slave medium.
Analog information from a master medium is reproduced and converted
to digital form. The digital data is stored in a digital storage
device. The digital data is recovered from the digital storage
device and recorded onto the slave medium at a rate that is higher
than the rate of recording information in analog form directly from
the master medium onto the slave medium. Accordingly, the time to
duplicate analog information from the master medium onto the slave
medium is reduced.
In U.S. Pat. No. 4,286,294 Nakauchi discloses a recording and
reproducing system for a recording medium where the recording and
reproducing head moves relative to the recording medium. Nakauchi's
system reduces wow and flutter that is produced when the recording
medium moves at a low speed. This is accomplished by generating a
sampling signal whose frequency varies depending upon the amount of
wow and flutter caused by the relative motion of the recording
medium.
In addition to the above references related to sound recording,
image processing systems exist that correlate related images to
reduce the noise in the images. For example, the "ZEISS".TM.
"DSM-960" scanning electron microscope can average multiple frames
to reduce the noise in an image. These .systems are typically
designed for subject matter that does not change between frames.
These systems generally do not have any means for correlating
subject matter from frames that are not aligned.
SUMMARY OF THE INVENTION
The present invention provides a method and apparatus for reducing
noise using a plurality of copies of a recording. The present
invention produces a master file with lower noise than the
available recording copies, and avoids the problems of losing
musical content caused by prior art pop and click removers. The
system comprises a recording playback unit, a computer system with
a sound input capability, and a high capacity storage system such
as a CD recorder.
In operation, a plurality of copies of a single recording are
played on the playback unit. For example, three records of an early
jazz recording. These recordings are digitized by the computer and
a separate recording file is formed for each copy of the recording.
The starting points of each of the recording files are then
aligned. The computer searches the recording file to find the first
instance of a music profile and defines that as a preliminary
marker for the start of the recording file. The computer or
operator selects one recording file to be a master recording based
on a preliminary determination of which file has the lowest noise
content. The other files are slave files.
To precisely align the recording files the computer evaluates the
alignment of a group of samples. In one embodiment, the standard
deviation of the difference between corresponding master and slave
samples is calculated using the preliminary identification of the
starting positions of the recording files as an initial alignment.
Corresponding samples are those samples located at the same point
in time in the respective recording copies as measured by the
current synchronized alignment of the recording files. The slave
file is then shifted in time relative to the master file and the
standard deviation of the difference of the re-aligned samples is
then computed. This process of shifting the files in time and
computing the standard deviation of the difference is then
repeated. This alignment process is repeated throughout the noise
reduction process at predetermined intervals.
The general principle of the noise reduction processes is that by
comparing the different recording copies the musical signals can be
distinguished from the noise, and the noise can then be reduced.
There are a large number of statistical processes that can be used
to reduce the noise in a recording using a plurality of recording
copies, including: averaging, calculating standard deviations,
calculating mode values, and performing threshold comparisons.
One embodiment of the noise reducing system of the present
invention uses an averaging process to reduce the level of white
noise. Because white noise is random the phase of the noise in
different recording files at any given sample location is likely to
be different. The computer uses this phase difference between the
noise components of corresponding samples in different recording
files to reduce the noise level by averaging the signals.
Numerous enhancements may be made to the averaging process to
further improve noise reduction. In one embodiment, as a
preliminary step, the system eliminates recording files that have
unacceptably poor sound quality relative to the other recording
files. At the start of the averaging process for a given sample,
the computer obtains the value of the selected sample from each of
the recording files. The computer then calculates the average of
those values. The computer may next eliminate sample values that
exceed a predetermined threshold difference from the average value.
Alternatively, the computer may eliminate the recording file, or a
predetermined number of recording files, that differ the most from
the average. In one embodiment, where there is a large number of
recording files, for example over ten, the computer determines the
threshold by calculating the standard deviation of the error
between each recording sample and the average signal level. The
computer then eliminates the recording files with an error greater
than a predetermined multiple of the standard deviation.
Alternatively, the computer may eliminate unacceptably noisy
recording files based on the standard deviation of the difference
between the master and slave samples. In another enhancement to the
averaging process, the computer approximates the signal-to-noise
ratio based on general properties of white noise, pops and clicks.
The computer may then eliminate any recording files that have a
signal-to-noise ratio below a given threshold. Also if samples from
particular recording files are repeatedly found to have
unacceptably high noise levels, the computer may eliminate the
entire recording file from the process. A further enhancement to
the averaging process is eliminating a predetermined number of
samples after a pop or click is detected.
This process of eliminating recording file samples based on
variations between recording file sample values reduces pops and
clicks as well as white noise. In contrast to prior art pop and
click removal systems, the present invention noise reducing system
avoids the problem of losing musical content when a pop or click is
eliminated because the present invention uses the samples from the
other recording copies for the output file in place of the
recording file samples corrupted by pop and click noise.
BRIEF DESCRIPTION OF THE DRAWINGS
FIG. 1 illustrates a system level block diagram of one embodiment
of the apparatus of the present invention.
FIG. 2 illustrates a noise free frequency spectrum and a frequency
spectrum with noise.
FIG. 3 illustrates time shifting files in a synchronization process
of the present invention. FIG. 3A illustrates the master file. FIG.
3B illustrates the slave file. FIG. 3C illustrates the master minus
the slave file. FIG. 3D illustrates the master minus the slave file
with an alignment difference of two samples.
FIG. 4 illustrates a noise averaging process of the present
invention. FIG. 4A illustrates an ideal signal. FIG. 4B illustrates
a first recording. FIG. 4C illustrates a second recording. FIG. 4D
illustrates an average of the first and second recordings.
FIG. 5 illustrates a flow chart overview of the noise reduction
process of the present invention.
FIG. 6 illustrates a composite time domain plot of a musical
recording.
DETAILED DESCRIPTION OF THE INVENTION
The present invention is directed to a method and apparatus for
reducing noise using a plurality of recording copies. In the
following description, numerous specific details are set forth in
order to provide a more thorough understanding of the present
invention. It will be apparent, however, to one skilled in the art,
that the present invention may be practiced without these specific
details. In other instances, well-known features have not been
described in detail in order not to unnecessarily obscure the
present invention.
The present invention produces a master file with lower noise than
the available recording copies, and avoids the problems of losing
musical content caused by prior art pop and click removers. The
system comprises a recording playback unit, a computer system with
a sound input capability, and a high capacity storage system such
as a CD recorder. In operation, a plurality of recording copies of
a single recording are played on the playback unit. These
recordings are digitized by the computer and a separate recording
file is formed for each copy of the recording. The recording files
are then synchronized. The samples from each of the recording files
are then averaged to reduce the noise components. A variety of
threshold comparison techniques can be employed to eliminate
samples and/or recording files that are outside of a computed range
for that sample based on the values of the master, the other slave
files or a combination thereof.
1. SYSTEM HARDWARE
FIG. 1 illustrates a system level block diagram of the present
invention noise reducing system. The output of playback unit 110 is
coupled to the input of sound card 116 in computer 114. Sound card
116 is coupled to hard disk drive 118 and compact disc recorder 112
through the bus in computer 114. Computer 114 is also coupled to
monitor 120 and keyboard 122.
In operation, first a user locates at least two copies of an old
recording. FIG. 5 shows a flow chart of the general steps in the
noise reduction process. The noise reducing system of the present
invention may be used with, for example, several old
phonograph-type records. The recordings are then played on playback
unit 110. Playback unit 110 may be any apparatus for converting
signals stored in a medium into electrical signals. For example,
playback unit 110 may be a turntable, reel-to-reel tape player,
eight-track tape player, cassette tape player, or compact disc
player.
The output signals of playback unit 110 are coupled to sound card
114. Sound card 114 digitizes the signals and creates a file for
each of the recording copies. "Recording copies" refers generally
to different copies of a recording made from one or more earlier
recordings, typically a master recording. For example, five
phonograph-type records of a music performance made from the same
original master recording of the performance. These recording files
are then stored either in hard disk 118 or in CD recorder 112. In
place of CD recorder 112, any one of a variety of long term storage
devices may be used including, for example, a laser disc recorder
or a removable tape cartridge drive. The files are then
synchronized and processed by computer 114 to produce a recording
with a reduced noise level, and a reduced number of clicks and
pops. This improved recording copy may then be used as a master to
make more copies, and the recording may be re-released.
2. SIGNALS AND NOISE
FIG. 2A illustrates an amplitude vs. frequency plot of an ideal
recording. In this ideal recording there is no noise or distortion,
and there is a linear frequency response throughout the audible
range. The absence of any frequency content between the peaks
indicates that there is no white noise in the recording.
An old analog recording is more likely to appear like FIG. 2B. FIG.
2B illustrates a recording with an undulating noise floor below
which the musical signals are lost. The noise floor is that
relatively flat frequency content that is interspersed throughout
the region for which signals are present. On an old recording some
of this noise was likely present on the recording when it was
originally produced. This noise component should generally be very
similar for each of the recording copies. Other noise components
are introduced into individual recording copies due to distortion
of the recording medium. For example, noise in old records may be
caused by embedded dust, imperfections in the vinyl or other
recording medium materials, and through degradation in the
recording grooves from playing the recording many times. Similarly,
tape recordings are also subject to embedded dust, and degradation
of the magnetic medium caused by friction including contact with
the roller and playback tape head. This component of the noise
varies between different recording copies.
FIG. 6 illustrates a time domain plot of a sample segment from a
music recording. The FIG. 6 signal is a composite of all of the
different frequency components present in the music. In one
embodiment, the present invention performs the noise reduction
signal processing steps on a digitized version of the FIG. 6 type
signal.
3. NOISE REDUCTION
The noise reduction process comprises several general steps. First
the recording copies are digitized into recording files. The
recording files are then synchronized. The samples from each of the
recording files are then combined to produce a composite value so
as to reduce the noise components. For example, in one embodiment
an average value is computed. A variety of threshold comparison
techniques can be employed to eliminate samples and/or recording
files that are outside of a computed range for that sample based on
the values of the master, the other slave files or a combination
thereof. These processes produce a master recording file with a
lower noise level than the recording copies.
a. DIGITIZING
To reduce the noise in recording copies a variety of signal
processing operations can be performed. First the recording copies
are played on playback unit 110. Playback unit 110 provides an
output signal to the input of sound card 116. For example, for a
record, a turntable is used for playback unit 110. The turntable
converts the contours of the grooves on the record into analog
electronic signals which sound card 116 converts into digital data
and stores as a recording file in hard disk 118 and or CD recorder
112. The signals are digitized at a high enough frequency to allow
computer 114 to shift the recording files in time one sample
backward or forward without the change being audible. This minimum
recording file resolution ensures that the recording flies can be
aligned with sufficient accuracy such that the files are temporally
aligned to an accuracy well within any difference that a human ear
could detect. In one embodiment, an 80 KHz sample rate is used.
This provides a sample rate of approximately four times greater
than the approximate maximum audible frequency of 20 KHz. Using
Nyquist's theorem, an 80 KHz sample rate is sufficient to digitize
frequencies up to 40 KHz. The user then repeats this process to
digitize and store the remaining recording copies that are
available. As will be discussed in more detail below, the more
recording copies that are used the more effectively the present
invention can reduce noise.
b. SYNCHRONIZING
After the recording copies are digitized, the first step in the
noise reduction process is to synchronize the recordings. The
computer must compare the same recording segment from each of the
recording copies in the noise reduction process in order to
accurately reduce the noise and to avoid averaging
non-corresponding samples of the recording files. This requires
that computer 114 align all of the recording files so that the
recordings start at the same time. In one embodiment, the system
determines the approximate start time of the music by using a noise
profile.
Phonograph-type records generally have unmodulated sections at the
start of the record and between songs or movements of the recorded
music. The sound level of these unmodulated sections typically is
the noise floor, aside from any pops or clicks that may be present.
The noise floor of a recording is that level of residual noise
present that is not dependent on a musical signal, as is
illustrated in FIG. 2B. In one embodiment, computer 114 therefore
determines the approximate start of the recorded sound by searching
for the first instance of a series of samples in the recording file
that exceed a predetermined threshold above the noise floor for a
predetermined time longer than an estimated duration of a pop or
click. Computer 114 then uses the synchronization processes defined
below to align the recording files to the desired accuracy.
Computer 114 should maintain the synchronization of the recording
files throughout the noise reduction process to compensate for any
drift, wow or flutter that may exist in the different recording
files. The term "wow" refers to slow pitch variation in the
recorded material. For example, wow can be caused in
phonograph-type records by an off-center hole. "Flutter" generally
refers to more rapid pitch variation that occurs in sound
reproduction as a result of undesired speed variations during the
recording, duplicating, or reproducing process. In one embodiment,
the recording file alignment in time should be within 25 .mu.S in
order to not affect the frequency response of the sound recording
when the samples are averaged. The synchronization process is
critical and potentially time consuming.
Playback unit 110 should be designed to maximize the accuracy and
consistency of the playback speed to facilitate synchronizing the
recording files. A turntable that plays back the recording copies
at a reduced speed may be used to improve the consistency of the
playback speed for the recording files. In one embodiment, computer
114 monitors the playback speed to ensure the speed is maintained
within a predetermined range. Approaches to reducing wow and
flutter are well known by those of ordinary skill in the art. For
example, U.S. Pat. No. 4,286,294, entitled "Recording/Reproducing
System" uses a reference frequency to remove varying frequency
components caused by wow and flutter, the disclosure of which is
hereby incorporated by reference.
To synchronize the recording files, computer 114 or the user
selects one recording copy to be a "master" recording. In one
embodiment, computer 114 selects the master by selecting the
highest quality recording based on a preliminary analysis of the
noise level in the recording, for example, the level of white
noise, or the number of clicks and pops. Alternatively, a user may
select the master recording file based upon, for example, listening
to the recording samples to determine which has the best sound
quality.
The recording copies are synchronized in blocks of either a
predetermined number of samples, for example 2000 samples, or a
variable number depending on an analysis of the relative drift of
the files. A variety of approaches to synchronizing the recording
files may be used. Recording files from different recording copies
vary due to the noise component of each recording copy, therefore,
computer 114 cannot synchronize the files by looking for an exact
match. One statistical approach to synchronizing a slave recording
is for computer 114 to measure the standard deviation of the error
between the amplitude of the master's samples and the slave's
samples over the selected block of samples. Standard deviation is
the positive square root of the expected value of the square of the
difference between a random variable and its mean. In this case the
random variable is the amplitude error, and the mean is zero. The
start time of the slave recording relative to the master recording
is then shifted, and the standard deviation measurements are
repeated. This process of determining a new shift value, measuring
the standard deviation for individual samples, and comparing the
calculated standard deviations to standard deviations for previous
shift values is repeated either a predetermined number of times or
until the standard deviation is below a threshold level. The shift
value with the minimum standard deviation is then used to align the
files.
As is well known by those of ordinary skill in the art, a variety
of search algorithms may be used to iteratively change the amount
that the slave recording is shifted so as to minimize the standard
deviations. One approach is to use predetermined shift values over
a certain range. For example, using sample values at ten sample
intervals over a range of .+-.100 samples. This process generates a
standard deviation table in which standard deviation varies as a
function of time shift values. The time shift corresponding to the
minimum standard deviation then correlates to the closest
approximation of the shift value required to synchronize the master
and slave recordings. The time shift-standard deviation search
process may be repeated until the desired level of synchronization
is achieved or until the master and slave files are synchronized to
within a single sample.
Computer 114 generally keeps the recording files synchronized
throughout the noise reduction process. The digitized noise in the
recording files may cause spurious variations in the table of
standard deviations as a function of time shift value. Computer 114
can take these two factors into account, and use the time shift
error from the standard deviation vs. time shift value table as the
error term in a digital filter controlling the speed at which the
computer moves through the slave file. Computer 114 does not make
any large instantaneous adjustments in the alignment of the
recording files. Computer 114 monitors the total drift between the
master and slave recording files and the standard deviation table
to determine if an error has occurred in the noise reduction
process. For example, if the system lost synchronization between
the master and slave recording files computer 114 would detect
large drift values and large standard deviation values. Computer
114 would then implement a corrective procedure, for example,
repeating the noise reduction process from the last point where the
recording files were known to be synchronized.
The synchronization process is illustrated in part by FIGS. 3A
through 3D. FIG. 3A illustrates an analog signal master. The curved
line in FIG. 3A represents the original analog signal. The
rectangles are the digital samples of the analog signal. FIG. 3B
illustrates a segment of the same signal from a slave recording. In
this case the slave recording file is one sample ahead of the
master recording file shown in FIG. 3A. FIG. 3C illustrates the
error between the signals shown in FIGS. 3A and 3B. FIG. 3D
illustrates the error between the master and slave recordings if
the slave recording of FIG. 3B is shifted another sample ahead of
the master in FIG. 3A. Thus, by varying the shift value and
measuring the corresponding error between the master and slave
files all of the files can be synchronized.
The frequency at which the synchronization process is repeated can
also be adjusted according to real time analysis of the number of
samples that can be processed before a sample shift error of
greater than one sample is detected. For example, if after 100
samples have been processed computer 114 performs the
synchronization process and determines that there has not been a
significant drift between the files, then computer 114 can increase
the number of samples before repeating the synchronization process,
for example 110 samples instead of 100. Similarly the number of
samples that are processed between synchronization procedures can
be decreased as needed where a shift of greater than a
predetermined size is detected, for example greater than one
sample.
c. SIGNAL PROCESSING
The general principle of the noise reduction processes is that by
comparing the different recording copies the musical signals can be
distinguished from the noise, and the noise can then be reduced or
eliminated. There are a large number of statistical processes that
can be used to reduce the noise in a recording using a plurality of
recording copies, including: averaging, calculating median values,
calculating standard deviations, and performing threshold
comparisons.
i. AVERAGING
One embodiment of the noise reducing system of the present
invention uses an averaging process to reduce the level of white
noise. Generally plotting a large number of measurements of a
single quantity subject to comparatively low level random noise
produces a group of values that form a bell curve, with the noise
free value of that quantity at the center of the bell curve.
Therefore for random noise in the recording files, the
corresponding sample values from a sufficient number of recording
files should also form a bell curve. White noise is random noise
that has a constant energy per unit bandwidth at every frequency in
the range of interest. Because white noise is random, the phase of
the noise in different recording files at any given sample location
is likely to be different. Computer 114 uses this phase difference
between the noise components of corresponding samples in different
recording files to reduce the noise level by averaging the
signals.
The averaging process is illustrated in FIG. 4. FIG. 4A illustrates
an amplitude vs. time plot for an ideal recording. To clearly
distinguish the signal from noise, a signal comprising a single
frequency is illustrated. FIG. 4B illustrates an example of a noisy
recording copy of the same signal. FIG. 4C illustrates a second
recording copy of the same signal. FIGS. 4A and 4B illustrate a
typical case where the noise levels on the signals are of
comparable peak-to-peak amplitude but have different phases.
Therefore, generally at any given time the amplitudes of the noise
signals are different. FIG. 4D illustrates an average of the
signals in FIGS. 4B and 4C. Averaging the noise signals generally
reduces the noise level as FIG. 4D illustrates. The reduction in
noise level is based on the fact that random noise is centered
around the original signal. When a plurality of recording files are
averaged the positive and negative deviations from the original
signal in the synchronized samples of the different corresponding
files at least partially cancel each other out. The resultant
signal converges on the original signal as the number of
independent recording files included in the averaging process is
increased.
ii. ENHANCEMENTS
Numerous enhancements may be made to the averaging process to
further improve noise reduction. One such enhancement is a
threshold level comparison step. In one embodiment, as a
preliminary step, the system eliminates recording files that have
unacceptably poor sound quality relative to the other recording
files. At the start of the averaging process for a given sample,
computer 114 obtains the value of the selected sample from each of
the recording files. Computer 114 then calculates the average of
those values. Computer 114 may next eliminate sample values that
exceed a predetermined threshold difference from the average value.
Alternatively, computer 114 may eliminate the sample, or a
predetermined number of recording samples, that differ the most
from the average. For example, for each set of corresponding
samples, the highest and lowest values can be discarded, under the
theory that they deviate so much from the average value that they
are not representative of the original sound value. In one
embodiment, where there is a large number of recording files, for
example over ten, computer 114 determines the threshold by
calculating the standard deviation of the error between each
recording sample and the average signal level. Computer 114 then
eliminates the recording file samples with an error greater than a
predetermined multiple of the standard deviation.
Alternatively, computer 114 may eliminate unacceptably noisy
recording files based on the standard deviation of the difference
between the master and slave samples. In the synchronization
process, computer 114 calculates the standard deviation between the
master and the slave recording files. The sum of the standard
deviations for each sample, calculated at the time shift value that
produced the minimum total standard deviation, is an indication of
the noise differential between the master and slave recording
files. Thus, computer 114 may eliminate recording files whose
minimum total standard deviation exceeds the minimum total standard
deviation of the other slave recording files by a predetermined
multiple of the standard deviation. In a further alternative,
computer 114 may eliminate individual samples where they differ by
greater than a predetermined amount from the corresponding master
sample value. Similarly, where a sample or group of samples is
determined to be sufficiently noise and distortion free this sample
or group of samples can be transferred directly into the output
file.
In another enhancement to the averaging process, computer 114
approximates the signal-to-noise ratio based on general properties
of white noise, pops and clicks. Computer 114 may then eliminate
any recording files that have a signal-to-noise ratio below a given
threshold.
Also if samples from particular recording files are repeatedly
found to have unacceptably high noise levels, computer 114 may
eliminate the entire recording file from the process. Eliminating
an entire recording file increases the speed of the noise reduction
process, and where there are a sufficient number of superior
recording files eliminating distinctly inferior recording files
further reduces the noise level in the final output recording
file.
A further enhancement to the averaging process is eliminating a
predetermined number of samples after a pop or click is detected. A
parameter is defined that approximates the duration of a pop or
click. This parameter can be set to a default value generally
representative of pops and clicks. Alternatively, a particular
recording file could be characterized to evaluate the typical
duration of signal features that exceed a threshold maximum value
associated with the music signal. In the averaging process when a
value is detected that exceeds a pop or click threshold level then
the following samples up to the predefined duration are deleted
irrespective of the value of the subsequent samples. The subsequent
samples are deleted on the theory that even if their values are
less than the threshold value the signal is likely to be due to
some dust or other contaminant on the disc than it is to be due to
a musical signal.
This process of eliminating recording file samples based on
variations between recording file sample values reduces pops and
clicks as well as white noise. In contrast to prior art pop and
click removal systems, the present invention noise reducing system
avoids the problem of losing musical content when a pop or click is
eliminated because the present invention uses the samples from the
other recording copies for the output file, in place of the
recording file samples corrupted by pop and click noise. The pops
and clicks generally occur at different times on different copies
of a single original recording. This is caused in part by
differences in the distribution of dust and other undesired
particles that have become embedded into the recording medium, for
example a vinyl type record. Therefore the samples dominated by pop
or click noise can be eliminated and replaced with the result of
the noise reduction process performed on the corresponding samples
from the other recording files.
The noise reduction process may be further enhanced by selectively
applying the above techniques. In one embodiment, computer 114
evaluates the distribution of the samples representing the same
time interval in each of the recording files. Based on the
distribution of these samples computer 114 determines which signal
processing techniques should be applied for this group of samples.
For example, where all of the sample values are within a certain
predetermined range, computer 114 may not eliminate any of the
samples. Similarly, where a large number of samples are within a
small range of values and a substantially smaller group of samples
vary widely, computer 114 may process the samples based on a mode
value approach rather than an average of the values, where the mode
value is defined as the most frequently occurring value, or range
of values, of the set of samples. Using the mode value eliminates
the distorting effect of sample values that widely differ from the
apparent original value as indicated by a large group of samples
distributed about a particular value.
In monophonic phonograph-type recordings both sides of the
recording grooves ideally form the same contours, and represent the
same musical signals. Stereo turntables read the inside and outside
contours of a recording groove as separate channels. Therefore, if
a stereo turntable is used for playback unit 110, the number of
recording files can effectively be doubled. Any samples of clicks
and pops in these recording files can be replaced with the
corresponding click or pop free samples from the other channel. The
channel with the click or pop is identified using the techniques
described above.
iii. POPS AND CLICKS ON A MASTER RECORDING
The present invention noise reduction process may also serve as a
process to prepare recordings for conventional pop and click
removal techniques that operate on only a single recording copy.
Furthermore, digital equivalents of conventional pop and click
removal techniques can be incorporated into the present invention.
Applying the present invention to reduce the noise content of the
recording copies enables the pop and click removal techniques to be
more effective because the pops and clicks are more clearly defined
when the noise is reduced. Conventional pop and click removal
techniques that operate on a single recording copy may be
particularly useful for eliminating pops and clicks that were
present on a no longer available original master. Pops and clicks
present on the original master are likely also to be present on all
of the recording copies.
d. DIRECTION OF PLAYBACK
Playing recording media in a direction other than the direction
typically used to play the recording can produce a higher quality
output signal. In one embodiment, phonograph-type records are
played backwards during the digitization process. Playing the
phonograph-type records using the reverse rotational direction
favors the side of the groove that has been exposed to less
resistance over the life of the record, as opposed to the side of
the groove that is exposed to greater pressure when the record is
being played in the forward direction. The technique of playing
recording media in the reverse direction can also be applied to
recordings on tape media.
Thus a method and apparatus for reducing noise using a plurality of
recording copies have been described. Although the present
invention has been described with respect to certain specific
embodiments, it will be clear to those skilled in the art that the
inventive features of the present invention are applicable to other
embodiments as well, all of which are intended to fall within the
scope of the present invention.
* * * * *